neuralnet-visualize 0.2.3

Generate a neural network architecture Image

Neural Network Visualizer

General Description

A module which creates a neural network image with the given architecture. It's a handy tool to see how your network is built as compared to a model summary.

Installation

Before installation

Before installing the module, run the below command at your prompt to install the graphviz

$ sudo apt install graphviz

Normal installation

$ sudo pip3 install neuralnet-visualize

Development installation

$ git clone https://github.com/AnuragAnalog/nn_visualize.git
$ cd nn_visualize

After installation

After installing the module, if you want to upgrade the module, run the below command.

sudo pip3 install neuralnet-visualize --upgrade

Future Works

• Add Convolutional layers, Maxpooling, Flatten layers
• Add Sequence model layers
• Directly from the pickle files
• Specific colors for activation functions
• Directly convert from pytorch models

Neural Network Visualizer Version History

0.2.3

• Added from_pytorch method
• Fixed a bug of multiple titles in an image
• Changed the logic of from_tensorflow method
• Fixed a bug, unique layer names for non trainable layers

0.2.2

• Added more docstring
• Can now be downloaded in py2

0.2.1

• Added title to the image
• Optimized some code

0.2.0

• Made it independent of tensorflow
• Fixed some bugs

0.1.4

• Fixed some bugs
• Added more layers maxpooling, avgpooling, flatten
• changed the shape of conv2d layer

0.1.3

• Added class docstrings with examples
• Added some more parameters for conv layer
• Disabled custom addition of layers after from_tensorflow call

0.1.2

• Added Conv layer, with kernel size parameter
• Refactored add_layer function
• Added Conv layer to from_tensorflow

0.1.1

• Added some possible exceptions
• Restructred basic functions in while initial import

0.1.0

• Added documentation

0.0.4

• Refactored summarize function
• Added a code part which colors output layer to red
• Added from_tensorflow method

0.0.3

• Added more display orientations
• Refactored the build network and add layer code
• Added some more file extensions to which image can be saved
• Added summarize function

0.0.2

• Added colors to type of layers
• Added Dense layers

0.0.1

Intial Version

• Starting code